Study Tips Normal Distribution Statistics Formulas 11 min read Updated November 15, 2026
BY: Minsa A. (Senior Statistics Editor)
Reviewed By: Statistics Fundamentals Team

The 68-95-99.7 Rule (Empirical Rule): Definition, Formula, and Examples

The 68-95-99.7 rule (also called the empirical rule) states that for data following a normal distribution, approximately 68% of values fall within 1 standard deviation of the mean, 95% fall within 2 standard deviations, and 99.7% fall within 3 standard deviations. The rule is a quick tool for understanding how data spreads around the mean without calculating every individual probability.

What Is the Rule? ↓ The Formula ↓ Worked Examples ↓ When It Applies ↓ Z-Scores ↓ FAQ ↓

What Is the 68-95-99.7 Rule?

The rule goes by several names depending on the context. In statistics textbooks it usually appears as the empirical rule. In engineering and quality control, the same idea is often called the three-sigma rule or 3-sigma rule. All three names refer to the same relationship between the normal distribution and its standard deviations.

For any dataset that is approximately normally distributed — meaning the histogram is bell-shaped, roughly symmetric, and has a single peak — these percentages hold:

Band Interval Approx. % Exact %
±1 standard deviation μ − σ to μ + σ 68% 68.27%
±2 standard deviations μ − 2σ to μ + 2σ 95% 95.45%
±3 standard deviations μ − 3σ to μ + 3σ 99.7% 99.73%

The symbols here: μ (the Greek letter mu) represents the population mean, and σ (sigma) represents the population standard deviation. These are population parameters, not sample statistics. If you're working with a sample, the equivalent notation is x̄ (sample mean) and s (sample standard deviation), but the rule's percentages still apply as long as the distribution is approximately normal.

The word "empirical" in this context means the percentages were derived from observing properties of the normal distribution. They are approximations of exact values computed from the standard normal distribution's probability density function. The rule rounds those exact values for practical use — a point worth returning to in the formula section below.

An introduction to the shape that makes this rule possible is on the normal distribution page.

The 68-95-99.7 Rule Formula

The Three Bands

The 68-95-99.7 Rule
P(μ − kσ ≤ X ≤ μ + kσ)
k = 1 → 68% (68.27%) k = 2 → 95% (95.45%) k = 3 → 99.7% (99.73%)

To use the formula, substitute your dataset's mean and standard deviation. The three intervals are:

  • About 68% of values fall in [μ − σ, μ + σ]
  • About 95% of values fall in [μ − 2σ, μ + 2σ]
  • About 99.7% of values fall in [μ − 3σ, μ + 3σ]
ℹ️
Accuracy note — exact values vs. rounded approximations

The rule rounds 68.27% to 68, 95.45% to 95, and 99.73% to 99.7. These are approximations. The difference matters most for the 95% case: 95% confidence intervals in statistics use ±1.96σ — not exactly ±2σ — to achieve precisely 95.00% coverage. The rule's ±2σ gives 95.45%, slightly more than 95%.

The Per-Band Breakdown (34-13.5-2.35 Rule)

Most explanations stop at the symmetric (cumulative) percentages. That is not enough for exam problems, which frequently ask about the percentage of values in a specific segment — not the whole symmetric band.

When the full distribution is split into one-standard-deviation bands on each side, the picture looks like this:

-3σ -2σ -1σ μ +1σ +2σ +3σ 34% 34% 13.5% 13.5% 2.35% 2.35% 0.15% 0.15% 68% 95% 99.7% within ±3σ
Bell curve showing the 68-95-99.7 rule with per-band percentages labeled. Each colored band shows the percentage of data in that one-standard-deviation segment.

Here is the same information as a table, which is what you need when solving problems:

Band (one side) Segment % in this band
Within ±1σ (combined) μ−σ to μ+σ 68%
μ to μ+σ (or μ−σ to μ) Half of ±1σ band 34% each side
Between ±1σ and ±2σ μ+σ to μ+2σ (or mirror) 13.5% each side
Between ±2σ and ±3σ μ+2σ to μ+3σ (or mirror) 2.35% each side
Beyond ±3σ Beyond μ+3σ (or mirror) 0.15% each side

Notice those four percentages per side: 34 + 13.5 + 2.35 + 0.15 = 50. Each side adds up to exactly half the distribution, which checks out.

Tail percentages are what students need for "greater than" or "less than" questions:

Threshold % of data above % of data below
μ + σ (or μ − σ) 16% 16%
μ + 2σ (or μ − 2σ) 2.5% 2.5%
μ + 3σ (or μ − 3σ) 0.15% 0.15%
⚠️
95% maps to ±2σ — never ±3σ

A common textbook error assigns 95% to three standard deviations. That is wrong. 95% corresponds to ±2σ. Three standard deviations cover 99.7%. Double-check any summary table you read; this mistake shows up in published study materials.

Step-by-Step Examples

How to apply the rule in four steps:

  1. 1Confirm the data is approximately normally distributed.
  2. 2Find the mean (μ) and standard deviation (σ).
  3. 3Calculate the intervals: μ ± σ, μ ± 2σ, μ ± 3σ.
  4. 4Interpret using the 68%, 95%, 99.7% figures — or the per-band table for segment questions.

Example 1 — IQ Scores (Standard Application)

Example 1 — IQ Scores

IQ scores are approximately normally distributed with μ = 100 and σ = 15.

Find the intervals covering 68%, 95%, and 99.7% of IQ scores. Then answer two probability questions.

1

±1σ: 100 − 15 = 85 to 100 + 15 = 115. About 68% of people score between 85 and 115.

2

±2σ: 100 − 30 = 70 to 100 + 30 = 130. About 95% of people score between 70 and 130.

3

±3σ: 100 − 45 = 55 to 100 + 45 = 145. About 99.7% of people score between 55 and 145.

4

Tail question: What % of people score above 130?
130 = μ + 2σ → data above μ + 2σ = 2.5%.

5

Band question: What % of people score between 85 and 100?
85 = μ − σ; 100 = μ. This is the left half of the ±1σ band → 34%.

✓ 68% score between 85–115; 95% between 70–130; 99.7% between 55–145. Above 130: 2.5%. Between 85 and 100: 34%.

Example 2 — Manufacturing Defects (Tail Percentages)

Example 2 — Manufacturing

A machine produces bolts with μ = 10mm and σ = 0.1mm. Acceptable range: 9.7mm to 10.3mm.

What percentage of bolts are within spec? What is the defect rate?

1

Identify the interval in σ terms.
9.7mm = 10 − 3(0.1) = μ − 3σ. 10.3mm = 10 + 3(0.1) = μ + 3σ. The acceptable range spans ±3σ.

2

Apply the 99.7% rule. About 99.7% of bolts fall within spec.

3

Defect rate. The remaining 0.3% fall outside the acceptable range — about 3 bolts per 1,000.

✓ 99.7% of bolts are within spec. Defect rate: approximately 0.3% (3 per thousand). Note: Six Sigma quality control aims for ±6σ limits, reducing the defect rate to roughly 3.4 per million.

Example 3 — AP Statistics Band Problem

Example 3 — AP Statistics

Heights of adult women are approximately normal: μ = 64 inches, σ = 2.5 inches.

Calculate all band boundaries, then answer one band question and one tail question.

1

Calculate band boundaries:

μ − 3σ = 56.5 in  |  μ − 2σ = 59 in  |  μ − σ = 61.5 in  |  μ = 64 in  |  μ + σ = 66.5 in  |  μ + 2σ = 69 in  |  μ + 3σ = 71.5 in

2

Band question: What % of women are between 61.5 in and 64 in?
61.5 in = μ − σ; 64 in = μ. This is the left half of the ±1σ band → 34%.

3

Tail question: What % of women are taller than 69 in?
69 in = μ + 2σ → data above μ + 2σ → 2.5%.

✓ Between 61.5–64 in: 34%. Taller than 69 in: 2.5%.

Need exact probabilities for non-integer z values? Use the z-score calculator.

Z-Score Calculator →

When Does the 68-95-99.7 Rule Apply?

The requirement is straightforward: the data must be approximately normally distributed. That means the histogram should look roughly bell-shaped — symmetric, unimodal, with the bulk of values near the center and tails on each side.

A quick check: plot the data as a histogram. If it looks reasonably bell-shaped, the rule is likely valid. For a more formal test, see the normality tests page or the guide on how to check normality.

The rule does not apply to:

  • Skewed distributions — income data is a classic example, where most values cluster near the bottom while a small number of very high values pull the right tail out.
  • Bimodal distributions — data with two peaks (two distinct subgroups) violates the unimodal requirement.
  • Heavy-tailed distributions — financial returns, for instance, have more extreme values than a normal distribution predicts. Applying the empirical rule here understates tail risk.

When Normality Cannot Be Confirmed: Chebyshev's Theorem

When the distribution's shape is unknown or non-normal, Chebyshev's theorem gives a lower bound on how much data falls within a given range. For any distribution with finite variance and any real k > 1:

Chebyshev's Theorem
P(|X − μ| < kσ) ≥ 1 − 1/k²
k = 2 → at least 75% within ±2σ k = 3 → at least 88.9% within ±3σ

These are weaker guarantees than the empirical rule (75% vs. 95%, 88.9% vs. 99.7%), but they hold for any distribution shape. Note that k can be any real number greater than 1, not only whole numbers.

±2σ ±3σ Requires normality?
Empirical rule (68-95-99.7) ~95% ~99.7% Yes
Chebyshev's theorem At least 75% At least 88.9% No — any distribution

Use the empirical rule when normality is confirmed; use Chebyshev's theorem when it is not. The Chebyshev's theorem calculator handles the arithmetic.

The 68-95-99.7 Rule and Z-Scores

The rule is naturally expressed in z-score terms. A z-score converts any normally distributed value to its distance from the mean in standard deviation units:

Z-Score Formula
z = (x − μ) / σ
|z| ≤ 1 → 68% of the distribution |z| ≤ 2 → 95% of the distribution |z| ≤ 3 → 99.7% of the distribution

The z-scores ±1, ±2, and ±3 are the thresholds the rule describes. What looks like a statement about raw data values (μ ± kσ) is equivalent to saying that z-scores between −k and +k capture a known percentage of a normal distribution.

Here's a concrete example. On a test where μ = 65 and σ = 10, a score of 75 gives:

z = (75 − 65) / 10 = 1

This score sits 1 standard deviation above the mean, placing it at the 84th percentile (50% below the mean + 34% in the [μ, μ+σ] band = 84% of scores fall at or below it).

Z-scores become essential when you need probabilities at points that are not exactly 1, 2, or 3 standard deviations from the mean. For those, the z-table gives exact cumulative probabilities. The empirical rule is an approximation good enough for integer multiples of σ; the z-table handles everything else.

Looking up probabilities for non-integer z values? The z-table has exact values.

View the Z-Table →

Real-World Applications

Standardized testing. IQ tests are designed with μ = 100 and σ = 15. SAT scores (historically) used μ = 500, σ = 100 per section. The rule immediately frames what "above average" and "exceptional" mean in context — a score above 130 (μ + 2σ) puts someone in roughly the top 2.5%.

Healthcare. Blood pressure, birth weight, and adult height are approximately normal, which is why clinicians talk about "within normal range." A measurement within ±2σ of a healthy population's mean covers about 95% of that population — a natural baseline for what counts as typical.

Manufacturing and quality control. Control charts use ±3σ limits, so that 99.7% of measurements should fall within bounds when a process is running normally. A data point outside ±3σ signals a likely process problem worth investigating. The ambition behind Six Sigma is to push spec limits to ±6σ, reducing the defect rate to roughly 3.4 per million opportunities.

Finance. Financial returns are often approximated as normally distributed over short time horizons. The empirical rule offers a first estimate of how often large swings should occur — though actual returns have heavier tails than a normal distribution, meaning extreme events happen more often than the rule suggests. Treat it as a starting point, not a guarantee.

Frequently Asked Questions

What is the 68-95-99.7 rule? ▼

The 68-95-99.7 rule (also called the empirical rule) states that for data following a normal distribution, approximately 68% of values fall within 1 standard deviation of the mean, approximately 95% fall within 2 standard deviations, and approximately 99.7% fall within 3 standard deviations. The exact values are 68.27%, 95.45%, and 99.73%.

What percentage of data falls within 2 standard deviations? ▼

Approximately 95% of data falls within 2 standard deviations of the mean in a normal distribution (exact value: 95.45%). This means roughly 2.5% of data lies more than 2 standard deviations above the mean, and another 2.5% lies more than 2 standard deviations below it. The 95% figure is why confidence intervals in statistics are often set at the 95% level — though those CIs use ±1.96σ, not exactly ±2σ.

Does the 68-95-99.7 rule work for all distributions? ▼

No. The rule applies only to approximately normally distributed data (bell-shaped, symmetric, unimodal). For skewed, bimodal, or heavy-tailed distributions, the percentages will differ. For any distribution (regardless of shape), Chebyshev's inequality provides a weaker but universal guarantee: at least 75% of data falls within 2 standard deviations and at least 88.9% within 3 standard deviations.

What are the exact values for the 68 and 95 percentages? ▼

The exact values are 68.27% (within ±1σ), 95.45% (within ±2σ), and 99.73% (within ±3σ). The rule rounds these for practical use. The 95% case is especially important: 95% confidence intervals in statistics use ±1.96σ (not exactly ±2σ) to give precisely 95.00% coverage. The rule's ±2σ gives 95.45%, slightly more than 95%.

How do you use the 68-95-99.7 rule step by step? ▼

Step 1: Confirm your data is approximately normally distributed. Step 2: Find the mean (μ) and standard deviation (σ). Step 3: Calculate the intervals: μ ± σ for 68%, μ ± 2σ for 95%, μ ± 3σ for 99.7%. Step 4: Interpret — about 68% of values fall in the first interval, 95% in the second, 99.7% in the third. For tail or band-specific probabilities, use the breakdown: 34%, 13.5%, 2.35%, and 0.15% per side.

What is the difference between the empirical rule and Chebyshev's theorem? ▼

The empirical rule (68-95-99.7 rule) applies specifically to normally distributed data and gives precise percentages: 68%, 95%, 99.7%. Chebyshev's theorem applies to any distribution with finite variance and gives minimum guarantees: at least 75% within 2 SD and at least 89% within 3 SD. Use the empirical rule when you have confirmed normality; use Chebyshev's theorem when you cannot.

Key Takeaways

📌

68%, 95%, 99.7% of data falls within ±1σ, ±2σ, ±3σ of the mean in a normal distribution.

📌

The per-band breakdown (34%, 13.5%, 2.35%, 0.15% per side) is what exam problems require for segment and tail questions.

📌

95% → ±2σ. Not ±3σ. Three sigma covers 99.7%.

📌

The exact values are 68.27%, 95.45%, 99.73%. The 95% CI in statistics uses ±1.96σ, not ±2σ, for precisely 95.00%.

📌

μ and σ are population parameters. Sample notation uses x̄ and s. The rule applies to either as long as the distribution is approximately normal.

📌

Chebyshev's theorem is the fallback when normality cannot be confirmed: at least 75% within ±2σ, at least 88.9% within ±3σ, for any distribution.

Once you have both the cumulative percentages and the per-band breakdown memorized, most normal-distribution problems on exams come down to identifying which band a range falls in. For everything else — exact probabilities at non-integer z values, confidence interval calculations, or practice problems with worked solutions — the statistics cheat sheet and statistics formulas page are worth bookmarking.